each of the studies. That is, the probabilities of getting statistically significant results if the true situations were the actual treatment differences and standard deviations found in the trials, were uniformly low.
In fact, the actual standard deviations in all 3 trials were between 10-12, certainly not around 7. In addition, the average treatment differences of study 203 and 205 were –2.75 and –1.5, respectively. The sponsor was unjustifiably optimistic to detect a difference of –2.5 in study 212 under these circumstances. Even if the treatment difference were realistic, the sponsor presumably knew that studies 203 and 205 produced standard deviations which would have required double the sample size as that required if assuming a standard deviation of 7. However, it is possible that they mistakenly used underestimates of standard deviations based on administrative interim analyses. In any case, study 212 was not adequately designed.
Discussion
Ironically, the only variable of the three major ones not to be statistically significant in the meta-analysis was the Depressed Mood item. This result must be seen in the light of the actual designs which were based on the total HAMD score, a composite endpoint which may be more sensitive than just one item. Ultimately, this submission may represent 3 nearly identically designed trials which appear underpowered to detect a small difference from placebo.
(signature)
David Hoberman, Ph. D.
Mathematical
Statistician
concur: Dr. Vevius (initialed; dated 2-14-95)
Dr. Dubey (initialed; dated 2-17-95)
Back a Page
Next Section - Chemist Review
Back to Wellbutrin SR® NDA Index Page
Back to Main
Index